Monitoring Refugee Camp Evolution and Population Dynamics in Dagahaley, Kenya, based on VHSR satellite data

The National Remote Sensing Committee (CNT), A Spatial Images End-Users Network in Madagascar: First Experience Return
8th Oct 2012
Development of an indicators system for monitoring climate change in the region of Marrakesh Tensift Al Haouz
8th Oct 2012

In the course of the severe drought at the Horn of Africa and the ongoing violent conflict in Somalia in summer 2011, more than 150,000 refugees arrived in Dadaab, Kenya, which is currently the world’s largest refugee camp complex. The enormous influx of people to the Dagahaley refugee camp, one of the three camps in Dadaab, brought the camp registration to a halt and revealed the need for a more efficient camp monitoring. Newly arrived refugees had to settle in the outskirts of the camp. The number and spatial distribution of dwellings could not be observed on the ground due to time and security constraints. In the frame of a Cooperation Agreement (Memorandum of Understanding, MoU) with Médecins Sans Frontières (MSF), the Centre for Geoinformatics at Salzburg University monitored the camp evolution using very high spatial resolution (VHSR) satellite imagery and provided in-depth information for supporting resource planning. Information on the amount and type of different dwelling structures and their spatial distribution was extracted by semi-automated analysis of WorldView-2 imagery (8 MS bands, 0.5 m GSD) from July 2011 and December 2011. Both images were partly affected by clouds and cloud shadows. Therefore, the eastern part of the December image was replaced by an additional image from January 2012.

The semi-automated dwelling extraction relied on object-based image analysis (OBIA), which provides a methodological framework for addressing complex information classes, defined by spectral, spatial, contextual as well as hierarchical properties. Expert knowledge is represented through rulesets coded in CNL (Cognition Network Language) in eCognition 8 software, which offers a modular programming environment for (image-)object handling. Objects may be addressed individually through class modeling, a cyclic process of segmentation and classification. For the analysis of the 1st timeslot three dwelling types were distinguished: tents, huts and dwellings with corrugated iron roof. Tents and makeshift huts could mainly be observed in the newly settled areas in the western outskirts of the camp, whereas dwellings with corrugated iron roof were the predominant dwelling type in the main part of the camp. The ruleset developed for the July image could be partly transferred to the December image. However, such clearly distinctive indicators of newly settled areas nearly have disappeared at the 2nd timeslot, e.g. only very few makeshift huts were still present and many dwellings with corrugated iron roof have been covered with white plastic sheeting due to the rainy season, which made a differentiation to white tents unfeasible. Therefore only one class dwelling was extracted for the 2nd timeslot. For shaded areas in both images, even though WorldView-2 still provided appropriate information due to its high radiometric resolution, the ruleset had to be slightly adapted to extract relevant objects. Finally, minor manual refinement was performed to eliminate obvious classification errors. The analysis of the July scene revealed about 23,400 dwellings: 13,950 dwellings with corrugated iron roof, 6,650 tents and 2,800 huts. In December 21,950 dwellings were extracted. In addition to single extracted dwellings the dwelling density (dwellings/km²) was calculated using Kernel density methods to provide easy to grasp information about the spatial distribution of dwellings. Based on the dwelling density the camp extent was derived automatically (see Fig. 1). A change analysis of dwellings aggregated on hexagonal units shows a decrease of dwellings in the western outskirts of the camp from July 2011 to December 2011. On the other hand, dwelling density increased in the main part of the camp and a minor increase of single dwellings in the eastern outskirts of the camp could be observed as well (see Fig. 1). Areas which were covered by clouds in either of the two images were excluded from the change analysis. Results have been delivered as maps in PDF-format as well as Google’s kml-files.

Figure 1: Change detection analysis based on single extracted dwellings in the Dagahaley refugee camp between July 2011 and December 2011. Blue tones indicate areas of dwelling decrease, red tones show an increase of dwellings and grey areas did not undergo a significant change. Clouds in either of the images were not taken into account for the change analysis (dashed areas). The camp extent of July 2011 is displayed in green, whereas the red outline shows the camp extent of December 2011. The WorldView-2 image in the background is a combination of the December 2011 image and the January 2012 image (eastern part) and is displayed in true colour composite.
The study shows that relevant and up-to date information in regard to amount and spatial distribution of affected population during humanitarian crises can be provided for inaccessible areas by making use of VHSR satellite imagery. Geo-information can contribute to make humanitarian aid more efficient, timely and effective.
This post was written by Petra Füreder, Daniel Hölbling, Dirk Tiede, Peter Zeil and Stefan Lang (Centre for Geoinformatics, University of Salzburg). Contact Petra Füreder at for more information.